Cropland productivity index (CPI) in China, 2001_2020, 250m

<b>1. Overview</b>This dataset provides annual estimates of the <b>cropland productivity index (CPI</b><b>) </b>for stable cropland areas in China from 2001 to 2020. The data is provided at a 250-meter spatial resolution.The data was created to overcome the limitations of traditional productivity assessments, which often lack spatial detail or temporal frequency. It is based on a remote sensing model called the <b>Crop Growth Index (CGI)</b>, which uses satellite data to monitor crop growth dynamics over time.<b>2. Methodology</b><b>2.1 The Crop Growth Index (CGI) Model</b>The CGI is a custom index designed to represent cropland productivity. It is calculated using this formula:<b>CGI = L_cgs * EVI_mean</b><b>L_cgs:</b> The length (in days) of the "critical growth period" for crops.<b>EVI_mean:</b> The average Enhanced Vegetation Index (EVI) value during that period.The "critical growth period" is defined as the time when the EVI value is above a certain threshold. This threshold (75% of the 3-year moving average maximum EVI) was determined through a sensitivity analysis to achieve the best correlation with national grain yield statistics.<b>2.2 Data Production Workflow</b><b>Satellite Data:</b> The model uses the 250m EVI data from the MODIS (MOD13Q1.061) satellite product.<b>Cropland Mask:</b> A stable cropland map was created by fusing multiple land use datasets to ensure only consistent agricultural areas were analyzed.<b>Processing:</b> The entire workflow was implemented on the Google Earth Engine (GEE) platform. This included smoothing the raw EVI time-series data and calculating the annual CGI for every pixel from 2001 to 2020.<b>3. Accuracy Assessment</b>The accuracy of the CGI dataset was validated using two approaches:<b>3.1 Direct Validation</b>The CGI data was compared against ground-based measurements of annual gross primary productivity (AGPP) from 37 cropland monitoring sites across China.<b>Result:</b> The CGI showed a good correlation, explaining approximately <b>60%</b> of the variance (average R-squared = 0.592) in the ground-observed productivity.<b>3.2 Indirect Validation</b>The CGI was compared with five other public productivity datasets by assessing their consistency with national statistics and agricultural survey data at county, provincial, and national scales.<b>Result:</b> The CGI demonstrated the <b>best overall performance</b>, especially in tracking the year-to-year changes (temporal trends) in productivity at both provincial and national levels. Its spatial patterns were also highly consistent with validation data.<b>4. Key Findings</b><b>National Trend:</b> From 2001 to 2020, average cropland productivity in China increased at an annual rate of <b>2.11%</b>. The growth was significantly faster in the first decade (2001-2010) than in the second.<b>Regional Patterns:</b> The Northeast Plain showed the highest average productivity, while the Huang-Huai-Hai Plain and Loess Plateau experienced the fastest growth rates.<b>5. Limitations</b>The model uses a single national threshold, which may reduce accuracy in some local areas.The relationship between CGI and actual yield may be non-linear.The validation relies on linear regression, which is a simplification of complex real-world relationships.

1. **数据集概述**
本数据集提供了2001年至2020年中国稳定耕地区域的**耕地生产力指数（cropland productivity index, CPI）**年度估算值，数据空间分辨率为250米。
本数据集旨在弥补传统生产力评估方法的局限性——传统方法往往缺乏空间细节或时间频率维度的信息。其基于一款名为**作物生长指数（Crop Growth Index, CGI）**的遥感模型构建，该模型利用卫星数据实时监测作物生长动态。

2. **研究方法**
2.1 **作物生长指数（CGI）模型**
CGI是为表征耕地生产力而设计的定制化指数，计算公式如下：
**CGI = L_cgs * EVI_mean**
**L_cgs**：作物“关键生育期”的时长（单位：天）。
**EVI_mean**：该时段内**增强型植被指数（Enhanced Vegetation Index, EVI）**的平均值。
“关键生育期”被定义为EVI值高于特定阈值的时段。该阈值设定为3年移动平均最大EVI值的75%，通过敏感性分析确定，以实现与全国粮食产量统计数据的最优相关性。

2.2 **数据生产流程**
**卫星数据**：本模型采用MODIS（MOD13Q1.061）卫星产品提供的250米分辨率EVI数据。
**耕地掩膜**：通过融合多套土地利用数据集生成稳定耕地地图，确保仅对持续稳定的农业区域开展分析。
**数据处理**：整个工作流基于**谷歌地球引擎（Google Earth Engine, GEE）**平台实现，包括对原始EVI时间序列数据进行平滑处理，并计算2001年至2020年每个像元的年度CGI值。

3. **精度评估**
本数据集的精度通过两种方式进行验证：
3.1 **直接验证**
将CGI数据与全国37个耕地监测站点的**年度总初级生产力（annual gross primary productivity, AGPP）**地面实测值进行对比。
**结果**：CGI表现出良好的相关性，能够解释约60%的地面观测生产力方差（平均决定系数R²=0.592）。

3.2 **间接验证**
将CGI与其他5套公开生产力数据集进行对比，通过评估它们在县级、省级及国家级尺度下与国家统计数据和农业调查数据的一致性进行验证。
**结果**：CGI整体表现最优，尤其在追踪省级和国家级尺度下生产力的年度变化（时间趋势）方面性能突出，其空间分布格局也与验证数据高度一致。

4. **核心发现**
**全国趋势**：2001年至2020年，中国耕地平均生产力以2.11%的年增长率提升，前十年（2001-2010）的增长速度显著快于后十年。
**区域格局**：东北平原的平均生产力最高，而黄淮海平原与黄土高原的生产力增长速度最快。

5. **局限性**
本模型采用单一的全国阈值，可能会降低部分局部区域的计算精度；
CGI与实际产量之间的关系可能呈非线性；
验证过程采用线性回归方法，这是对复杂现实关系的简化处理。